### Table 7. Results for on-line framework tracking an applied goal.

"... In PAGE 94: ...2. Table7 shows the performance of our on-line adaptation framework, together with the onLineAdaptationTr and onLineAdaptationPower functions, for various goals applied to several DSP benchmarks. The starting schedule that is refined is found, as with the other experiments in this section, by using the standard critical path scheduling algorithm.... In PAGE 95: ...053), (T, 215), (P, 0.050), (T, 210), P} In Table7 , the column titled Goal represents the goal that is applied to the application. Also, for a non-negative integer , column denotes the value of a metric of the best schedule found by the on-line adaptation framework, after schedules have been assessed by executing them for some time.... In PAGE 95: ... Also, for a non-negative integer , column denotes the value of a metric of the best schedule found by the on-line adaptation framework, after schedules have been assessed by executing them for some time. For the same exper- iments, which are reported in Table7 , Table 8 shows the times at which different constraints associated with the applied goals, are satisfied. For a given goal that is applied on an application, for non-negative integers and , denotes the number of schedules that have been executed in order to assess them before the th con- straint in the applied goal is satisfied.... ..."

Cited by 1

### Table 1. Comparison to Gaussian RBF with on-line learning.

"... In PAGE 4: ... The selective forgetting algorithm was applied to compute the second layer weights. Table1 summarises the one step ahead prediction results. It can be seen that the proposed neuro-fuzzy learning algorithm has a similar performance to the RBF with 20 hidden units.... ..."

### Table A3.3 On-line sensor used for redox potential measurements Parameter Instrument Cell Instrument

2006

### Table 8. Results for on-line framework tracking an applied goal.

"... In PAGE 95: ... Also, for a non-negative integer , column denotes the value of a metric of the best schedule found by the on-line adaptation framework, after schedules have been assessed by executing them for some time. For the same exper- iments, which are reported in Table 7, Table8 shows the times at which different constraints associated with the applied goals, are satisfied. For a given goal that is applied on an application, for non-negative integers and , denotes the number of schedules that have been executed in order to assess them before the th con- straint in the applied goal is satisfied.... ..."

Cited by 1

### Table 1: Number of parameters to be updated on-line.

in Comparison of Different Growing Radial Basis Functions Algorithms for Control Systems Applications

"... In PAGE 3: ...3. NNs Architecture Comparison In Table1 a schematic comparison of the analyzed classes of SGNNs is shown. The comparison is made in terms of the number of parameters that the algorithm needs to update at each learning step.... ..."

### Table 2 Overview of sensors

"... In PAGE 3: ...ilter. Nitrification was not obtained. Integration of on-line sensors In the framework of the project additional on-line sensors were installed at the full-scale plant. An overview of the existing and additional sensors is given in Table2 . The data of all sensors were collected on a central PC.... ..."

### Table 2: Average values of relative error and number of epochs required for conver- gence for the on-line algorithms.

1994

"... In PAGE 23: ...igni cantly faster than the corresponding batch algorithm (cf. Figure 2). This is also true of the on-line HME algorithm, which has nearly converged within the rst epoch.The minimum values of relative error and the convergence times for both ar- chitectures are provided in Table2 . We also provide the corresponding values for a simulation of the on-line gradient algorithm for the HME architecture (Equation 17).... ..."

Cited by 612

### Table 1: On-line RWA algorithm for sub-path protection

2005

"... In PAGE 6: ... For sub-path protection, however, to the best knowledge of the authors no on-line algorithm has been published so far that works at wavelength granularity. Therefore, the algorithm in Table1 is used in the simulator to determine routing and wavelength assignment. Goodness of paths is evaluated as follows.... In PAGE 7: ... Computing the paths as described in [23] will not yield a closed DPM in general. As for the algorithm in Table1 there are two options to with respect to this problem. The rst one is that if a path segment is missing from the DPM the protection path allocation is considered to be unsuccessful at (**).... In PAGE 10: ...015 0.01 PL Unavailability LP SPP SubPP (d) Figure 4: (a) Blocking probability at = 200 as a function of NL and 1 Amin at = 200 as a function of (b) NC, (c) NL and (d) PL criterion used for evaluating the goodness of the solutions in the algorithm on Table1 is optimized for decreasing overall resource usage and blocking. This is done by means of assigning lower cost to solutions with more possible sharing, which, in turn, yields a lower availability.... ..."

Cited by 2

### Table 1. The On-Line Monte-Carlo GPTD Algorithm Parameters: ,

2005

"... In PAGE 5: ...erested reader to (Engel, 2005, Appendix A.2.3) for the complete derivation (with state dependent noise). In Table1 we present the resulting algorithm in pseu- docode. Some insight may be gained by noticing that the term rt 1 ~ k gt; t ~ t 1 appearing in the update for dt is a temporal di erence term.... In PAGE 5: ...emporal di erence term. From Eq. (3.14) and the def- inition of ~ kt (see Table1 ) we have rt 1 ~ k gt; t ~ t 1 = rt 1 + ^ vt 1(xt) ^ vt 1(xt 1). Consequently, dt may be viewed as a linear lter driven by the temporal dif- ferences.... ..."

Cited by 10

### Table 1. Lower bounds on the competitive ra- tio of on-line algorithms, depending on the platform type and on the objective function.

2006

"... In PAGE 3: ....1 while in another one (task T on P3) it will be, say, 1.2. Clearly, the minimum of the performance ratios over all ex- ecution scenarios is the desired bound on the competitive ratio of the algorithm: no algorithm can do better than this bound! Because we have three platform types (communication- homogeneous, computation-homogeneous, fully heteroge- neous) and three objective functions (makespan, max-flow, sum-flow), we have nine bounds to establish. Table1 sum- marizes the results, and shows the influence on the platform type on the difficulty of the problem. As expected, mixing both sources of heterogeneity (i.... In PAGE 13: ...ing algorithms. The major contribution of this paper lies on the theoretical side, and is well summarized by Table1 . We have provided a comprehensive set of lower bounds for the competitive ratio of any deterministic scheduling algorithm, for each source of heterogeneity and for each target objec- tive function.... ..."

Cited by 1